The Exchanges

Q So do you launch a very expensive taxpayer-funded manhunt if it turns out that it was just a fluke that comes back?

A We did. Yeah, I did get a question about this on Twitter from, um, someone, uh, this person, Joran, who was asking why taxpayers have to pay millions to rescue, uh, the people who took this huge risk aboard an unlicensed vessel. Like for instance, if you want to climb Mount Everest, you need to get a permit. So do you think it's like, I, this is a big point in the story that people have talked about, you know, it's very big government rescues, um, that were basically were sent out to people who did take a risk. And, you know, we have, and this, the other thing that's been talked about is there was this, you know, boat of migrants that went down off the coast of Greece, 800 people died, and you didn't see half of the government resources marshaled there. So talk about that.

AI assessment note: “We did. Yeah, I did get a question about this on Twitter”

Q how they're gonna, they have so much going on right now. They have to have so much going on. I, I said it last week, like the math still doesn't work to me of like, what does their cloud business look like? What does their consumer devices business actually look like? But they have to focus and I don't, I don't think they will, but. How do they see themselves?

A I don't think it's a marketing campaign. Here's a way I think that you can make the product better and not resort to sycophancy. Uh, improve memory. Memory to me within ChatGPT has been the killer feature. Uh, I find ChatGPT is remembering much more about me, uh, in a way that I find not creepy, very useful. I've been planning this trip coming up in December. Um, And we've been talking about logistics a tremendous amount. And, uh, and it, it, it really does a great job of pulling for, from past conversations, uh, and, and again, like losing this goldfish brain, uh, that so many chatbots have starting to get to know people better. Uh, I mean, does that make you more, make it more of a companion? Yeah, because it knows you. Uh, but I think that will make it, uh, unbelievably more useful, and that's the direction I think they're going.

AI assessment note: “Here's a way I think that you can make the product better”

Q minute nuanced debate about, like, where is anthropic now compared to, like, their starting ideology, but if you zoom out, what's obviously important here is that The ideology is the powerful driving motivator at the beginning of these companies, right? And then it becomes a question of like, how does that ideology like morph and live on or evolve as companies get bigger and have to deal with bigger problems?

A Yeah. And you've, you've been like, I'd say one of the foremost chroniclers of this, this notion that Companies can sort of, uh, in, in their, uh, companies, individuals, in their belief in this big idea can, can sometimes be blind to, um, you know, some of the simplest errors or simplest vulnerabilities that might take down the whole house of cards. Here is, um, One of my favorite stories that I've read, uh, the first couple paragraphs, um, that, that you wrote, uh, along with a co-author of Bloomberg. I wonder if you'd know where this is going. One of the most lavishly funded gadget startups in Silicon Valley last year was Juicero. It makes a juice machine. The product was an unlikely pick for top technology investors, but they were drawn to the idea of an internet connected device that transforms single serving packets of chopped fruits and vegetables into a refreshing And healthy beverage. Doug Evans, the company's founder, would compare himself with Steve Jobs in his pursuit of juicing perfection. He declared that his juice press wields four tons of force enough to lift two Teslas. Uh, ok, but after the product hit the market, some investors were surprised to discover a much cheaper alternative. You can squeeze the Juicero bags with your bare hands and get the same juice out. I mean, isn't that characteristic of this, like, the guy's calling himself Steve Jobs, believe in …

AI assessment note: “Here is, um, One of my favorite stories that I've read”

Q endeavor? At what point does OpenAI reject the idea of, you know, adding more users to its platform for the sake of its mission? Um, I think that it is, I, my take about the SOAR app is I don't think that OpenAI is setting itself up for success with this because they are opening themselves up to, you know, the, the kind of incentives of a social media app.

A Right. Yeah. You asked that question in your story. At what point does open AI's consumer business overtake its nonprofit mission? Uh, I, I think it's already there. So, uh, but anyway, we can leave that open since you left it open as, as a question in your story. I won't, you know, put you on the spot on that one. Um, but yeah, it's a valid, valid concern. And yeah, I think we can both agree that this is not just, uh, a video generation technology. It's a social app. And it does sort of then put forth this really interesting question, which is what's going to happen to the creator economy. I'm writing about this in big technology this week. It'll probably come out right before our, our episode comes out, but, uh, you have this, you know, creator economy is kind of a misnomer. There's only a small amount of people that are able to, um, work full-time creating content on the internet. That being said, uh, some of the people who have watched this Have said basically like this puts, uh, creators right in the crosshairs of AI. And in fact, I had a, I guest hosted the twit podcast over the weekend and I had a great quote from one of our panelists, Ari Paparo of MarketExture Media. He said, the creators are all high on the hog right now, but I would be thinking they're one of the first things that could be totally annihilated by AI. I think it's worth taking a minute to talk about th…

AI assessment note: “I think it's already there. So, uh, but anyway, we can leave that open”

Q What do I do with them? If, uh, if the AI note taker that Salesforce provides me or zoom provides me or Microsoft provides me, Is more useful and cost nothing. Like, what do I do? What do I do with that junior person if I don't need them in the meeting and how are they going to learn to be an advisor if they're not needed in the meeting?

A Yeah. So there might be just more extended training, but I think the, the, the second point that I wanted to make here, and it, by the way, like maybe that's too sunny. It probably is too sunny now that I'm saying it out loud. But the second point here is that It, you know, we do have declining birth rate in the U S and in the developing world, uh, developed world in general. Right. And so you're just going to have much fewer, uh, sort of people seeking jobs in those entry and those entry level fields than we have now. And, you know, you might need the AI. The reason why Japan has pushed so hard into AI automation, Um, sort of the, I think you even talk about this in your book, the factories that are just like lights out factories where you don't have anybody is because they needed it. Yeah. Like in Japan, they didn't have enough of those young people seeking jobs and in the U S and in China in particular, that's going to be a major issue as we look, 1020 years in the future.

AI assessment note: “we do have declining birth rate... you're just going to have much fewer”

Q Yeah, yeah, I think, I think provenance matters, you know, because it's provenance. Is this like a sleeper spy?

A So this is my sort of crackpot theory of the case, but I think that we are underestimating when we talk about humanoids, um, how violent human beings will get against them. We just saw in LA there were this, this burning of the Waymos. Um, you can look at that at a bunch of different ways. Uh, I think it's being underappreciated how That is, in some ways, a symbolic revolt against automation and big tech, and if, if you, let's say, okay, just putting this in, in a story context, you work in a factory. A humanoid robot comes in and takes your job. The next day you see a humanoid, let's say, delivery robot walking down the street. You don't, can't provide to your family anymore. You know that thing has cameras on it. You don't care. You're, you're, you're tipping it over at the very least.

AI assessment note: “So this is my sort of crackpot theory of the case”

Q uh, with Tesla being a, like, if it was a more progressive brand before the typical owner, at least was more progressive and environmentally focused. I don't see how you get that back. If they're to just kind of actually launch some just wildly industry leading products, I think that's interesting, but. I don't know. I, I think there's long-term damage that's gonna be tough to undo. What about you?

A Yeah, I definitely think there's going to be long-term damage at the companies and, but we'll see. I think that we tend, people tend to be quickly forgiving. I don't know. Not always when it comes to a business leader, uh, entering politics in this way. Uh, but to me, I think one of the interesting things is the way that this did come to an end. It was a fizzle, not a breakup with Trump. If you know what I'm saying, like a lot of people expected this to blow up, but Musk has also said that he plans to keep his small office. At the White House, Trump said he's gonna be involved, so I think that the thing that really did Musk's effort in was politics. Like, it turns out that politics is political, and there's a, you know, when you're trying to change a system the way that he tried to change it, there are certain people you have to win favor with, and he ended up sort of sparring with a lot of cabinet secretaries in a way that sort of ended up dooming Uh, the, the initiative. Now we're going to have Bill Vasson, who's the CTO of Booz Allen in a couple of weeks, and he's going to talk a little bit about the way that these Doge priorities live on in terms of getting the government to use technology to be more efficient. Uh, but ultimately, uh, the, the Doge era, Elon's Doge era fell far short of the goals that even he set. And I think that is just the bow you tie on it.

AI assessment note: “Uh, but to me, I think one of the interesting things is”

Q to Microsoft or went back to Amazon. And is it counting as revenue? Where is it counting as revenue? What's an investment? Like these questions we've been asking, It ha everyone has kind of been asking, but have just, has just let it go because no one's had to actually really dig into the numbers here or actually kind of be, have their feet held to the fire on them.

A It's hard. It's we, I think we're both bullish on the technology, but we question the economics and it would be one thing if the, you know, AI enterprise build out was going so well, That, you know, this number was logical, but it's not. Uh, enterprises are super slow. There's this MIT study, which has holes in it, which I think we should talk about more in terms of like the rigor of that study. Um, but that says that 95% of businesses are not getting profit, uh, on their AI investment. And you know what? It's, there, there has the, what's been interesting is less the study, but more the reaction. Where it's been just a lot of people nodding their heads saying, yeah, that feels right. As opposed to, this is so wrong. We are getting that ROI. We haven't heard the chorus come back in opposition to that study. And so when you're in that moment where you're like, okay, you have to do two trillion in revenue by 2030, we're about to hit New Year's 2026. And we're not seeing, we're not seeing the, the momentum I think that you would need to sort of justify this enthusiasm. That's where I think our job is to Pause and be like, what the hell is happening here?

AI assessment note: “we question the economics and it would be one thing if the, you know”

Q just say sell for now. Um, because this thing about the benchmarks, I, I personally don't fully understand the benchmarks, and what they're weighted against, and what they're sourced against, and, you know, is, are these completely standardized across the industry? Um, those are really impressive, impressive numbers, but what do they actually mean? And again, I, I think it's very hard to really judge this until you try it.

A Right. And I'll just give one more piece of evidence, uh, on my side. I know this is unfair, but, um, I, I just saw a thread with a stream of folks who've like tested, uh, a one on their personal benchmarks. Like a lot of people have their like own personal tests, uh, to give AI. And it's, it's just like person after person talking about how much better it is, uh, than their previous tests. Okay. But the other thing here is, is really like, Maybe, yeah, I don't know, like, this is preview, of course, and you, you, you mentioned it, and, and you're totally right that we've seen Sora, uh, which has been a lot of hype and not released yet, and we've seen the voice, uh, uh, or GPT-Foro voice application, and the status of that is, is still, you know, up in the air. And, you know, should we, uh, give OpenAI a couple weeks of gratitude for the magic intelligence? In the sky and have our, more of our toys soon? I don't think so.

AI assessment note: “folks who've like tested, uh, a one on their personal benchmarks”

Q the secret sauce of the tech giants has been decoded, right? I mean, Your book is, is, uh, is the best example of that. Many organizations use it. When I look at in the pocket, I saw so many cases coming back in how to reorganize our organization, how our governance looks like, and yet still at schools and universities, people are not taught about these principles. What, what's wrong?

A Uh, well, I think that I wrote this book not for the tech giants. I mean, if they want to read it, they can read it, but this is for everybody else to learn from their systems. And as I wrote it, uh, I really started to To really believe, uh, and I still, I still believe this now, that this stuff can most benefit, uh, government, uh, you know, nonprofits, education systems, uh, who need this sort of, uh, culture and, um, inventive mindset in order to, to better achieve their missions, right? We need these, these, these companies and these organizations, um, you know, to be successful, to better lift up society, And, you know, I mean, I worked for the government for two years at the start of my career. And let me tell you something. I mean, if the government worked on being more efficient and being more inventive, um, as opposed to what it exists now, which is, you know, unfortunately very bureaucratic and operating out of position of fear, um, we'd be much better off. They'd be better off. The employees would like it more and our society could benefit a lot too.

AI assessment note: “I wrote this book not for the tech giants... this stuff can most benefit, uh, government”

Q much more as a result and that everyone is trying to think about. How did they not just survive as a company into the AI era, but thrive? Like, how do you really get the advantages of the fact there's going to be so much more activity? And if you don't have AI in there, if you shut it out, then you're actually going to be declining, not thriving. Right.

A This kind of makes open AI puts open AI. So first of all, you're going to bring you talk about people using codecs. So one of your colleagues shared, and I think you've talked about this too, that you've brought chat GPT into codecs so you can bring codecs into ChatGPT, which is basically like if we're users of ChatGPT, this experience that we talked about of ChatGPT not only suggesting what you might want to do next, but going to do it for you, that's going to happen. And so it makes you effectively an operating system, don't you think? But not the operating system like an iOS, where you would like go open up the your phone and then tap different apps. It's almost as if all interaction with all apps will happen. Through this interface. Is that the ambition?

AI assessment note: “It's almost as if all interaction with all apps will happen. Through this interface.”

Q the media person in me. I love how it was outwardly said that X is going to promote articles and encourage people to write articles on it. And then we have coincidentally had our first gigantically viral X article that even ended with the Author on, on, uh, CNN. So, should we stick with Substack, guys, or is it time to go X only? Those are, that's where I'm starting.

A I, I, let me just say this. Your answer began with this idea that I accepted Matt Schumer's premise, and I don't know if I'm fully on board with what he's saying. In fact, I think there was a good amount of bullshit in his article. Now, there are certain parts of things that I do agree with. Um, but here's one thing that I thought he was completely wrong about, and he was talking about basically how the AI is improving itself, talking about this concept of recursive self-improvement. He writes, the AI labs made a deliberate choice. They focused on making AI great at writing code first because building AI requires lots of code. If AI can write that code, if AI can help build the next version of itself, a smarter version which writes better code which builds will build an even smarter version. Making AI great at coding was the strategy that unlocks everything else. He says they've now done it and they're moving on to everything else. Uh, this one, I, first of all, I'll turn to Steven and then, and then to Ranjan. I mean, this idea of recursive self-improvement, I would posit it's not here. It's not here. You know, are AI engineers using, um, you know, some AI tools for product testing? You know, maybe they are. But the idea that the actual brain of the model is being made smarter with the, with the actual model itself, um, you know, doesn't seem right to me. That, to me, felt lik…

AI assessment note: “I don't know if I'm fully on board with what he's saying.”

Q I think a lot of the answers, other than detecting the user and its humanity and its verified or valid identity, it's also trying to figure out that very question. There are billions of dollars invent invested in building digital experiences, dozens of billions of dollars every single year. Now, this is the question. How do you change it? How do you make it enabled for non-human users as well?

A I want to talk about that. And first, let's just talk about detection. We'll go through the path, detection, experience, and cybersecurity. Um, so let's talk about detection first. I was just, uh, doing a lot of research for a trip I have coming up where I needed to rent a car in Milwaukee and drop it off in St. Louis. And it was a pretty exhaustive search because I was like thinking, all right, am I going to fly into a different airport? Am I going to fly into these airports? What's the cheapest car? Um, and I'm going back and forth. I'm on kayak looking at flights. I'm at hot, I'm hot wire looking at cars. Then I'm back at kayak changing flights. I'm back at hot wire changing cars. Now I'm changing, looking at cars on kayak. I think this is like a pretty normal experience when it comes to travel booking. I would have loved a bot to go do that for me, and because my activity is multi-site, um, it's probably hard for those websites, I would imagine, to figure out Whether I'm the human me or the, let's say, open AI operator version of me going out and doing these searches. So how would a company even begin to try to detect the bot behavior in this situation?

AI assessment note: “I want to talk about that. And first, let's just talk about detection.”

Q of Barack Obama stepping on a scale and the model goes, knows that Barack Obama is like trying to Increase the sales like a joke. Um, but can models like really understand what's going on in these images to a much deeper level? Um, so there's large, uh, there's large vision models that are starting to incorporate language and images, but vision, there's no real all encompassing multitask vision model.

A So I have a couple of questions for you on this. Um, first of all, well, on the vision part, um, does that, does that sort of play into like a lot of people have been talking about how You know, one of the biggest applications of this current generation of AI is going to be an augmented reality, right? And like the meta has those glasses where like, you're not, you don't have an overlay right now, uh, but you can talk to things, uh, to their AI bot and it will look at the world and then give you a sense as to like what you're looking at, or you can even just ask questions about things and it will talk to you. So I'm curious, like how, how, um, How seriously you take this, this new era of augmented reality that we seem to be heading in because speaking of one of your tweets, you wrote, it's going to be hard to beat a computer in your pocket. You can use inconspicuously when you need to. So it sounds like you're a believer that the phone is going to be the, the way that we're going to interact with computing for a while, but, uh, maybe there's something I'm missing. Yeah.

AI assessment note: “So I have a couple of questions for you on this.”

Q do we maintain this? And if I'm a developer and I shipped an application and it goes, doesn't matter if I built it myself or through AI, but it goes down tomorrow and I'm not able to fix it and get it up again, I'm not doing my job, right? So this is a big question. How do we continue to own the application while leveraging increasing amount of automation?

A Okay, so we've talked about the liabilities and the tools you'd use to build AI, uh, liabilities in the infrastructure used to store key information for AI, and now liabilities in AI code and vibe coding in particular. But there's another side to this, which is, I love the fact that, you know, people are able to build things just by a prompt now, or maybe a little bit more sophisticated. They prompt, they code a little bit, they build off each other, and now something that might have taken a team of 10 can be done with one person. Um, but, you know, we've called them threat actors. I call them bad actors. Basically bad guys looking to do damage or looking to hack into, to, uh, computer programs. They have all these tools at their disposal as well. So are we already seeing them putting those tools to use in an attempt to hack into software? And has that increased the sophistication and the level of hacking that we're seeing already?

AI assessment note: “So are we already seeing them putting those tools to use in an attempt”

Q EA assumptions? Did it feel like we were buying into too many assumptions? Did it feel like a reasonable, you know, feel free to be a red team this and be harsh, but like, what was your sense on, uh, for somebody outside? I, I, uh, yeah, I'm assuming you're not necessarily in EA. What was your sense of like, uh, were we assuming too many things in that conversation?

A Yeah, no, I'm not EA. To me, it was, I, I don't, I don't, I don't think I had enough grounding to be able to, um, to really answer that question well. I thought it was interesting, like, I felt like it was a pretty, uh, I think EA's should, like, talk a little bit more, is my perspective, and I've tried to reach out to many of them, uh, especially when there was this, like, six-month pause, uh, that was funded by Open Philanthropy. They, they call for the six-month pause on AI. Uh, uh, development and was sort of like met with like a dismissive no on that front. So I think that's kind of where, uh, some of the skepticism comes from. Um, but I will tell you this because, uh, I think you saw the screenshot that, I uploaded five of your podcasts into Claude and was like, all right, tell me a little bit about what's important to Dworkesh. And then I was like, what did I say? I said, uh, basically like, do you think Dworkesh is, is an effective altruist? And Claude says, um, there's a high probability that Dworkesh Patel is an effective altruist, or at least strongly influenced by EA ideas. Based on the strong alignment between his express views and EA priorities, I would estimate the probability is quite high, perhaps in the range of 70 to 90%.

AI assessment note: “I don't think I had enough grounding to be able to, um, to really answer”

Q to see the same exact thing. As you get more comfortable letting co-work run, you start a task, and then you start a second task, and you move on, and you just do more in parallel. And I think there's just a lot of opportunity to make this experience very nice, and to make it more obvious for people. How, how do you do this? When do you do it?

A Right, and, and it probably extends to the way that you use a chatbot, right? And it's interesting, because Anthropix had this, um, kind of interesting relationship with the chatbot. Started out as technology first, decided to build the chatbot, ship Claude, and then just kind of moved more towards enterprise, like you looked at all the charts, and, um, and Claude was always at the bottom, but now you're seeing Claude's usage rise. And I have a thought, and I'd love to check this by you, that the future of the chatbot is Is not like I'm going to give you a question and you'll give me an answer. It's I will give you a question or, you know, talk to you about a problem and you, the chat bot will then suggest some sort of action you can take on my behalf. Like right now I'm talking a lot about a trip to India and what I think I'm going to get back in the future is this thing being like, like what you said, not having this like secondary step between having to go there and, and book the flights. A more proactive chatbot that's gonna say, okay, let me take the, let me take care of this for you. Is that the right direction? Like, am I thinking about that?

AI assessment note: “I have a thought, and I'd love to check this by you”

Q it's at least a little predictable and just, you know, it'll just affirm everything you say, that's, it's almost better versus the self-aware side of things. Uh, maybe that makes it a little spicier, makes it a little more unpredictable. Maybe does that make it more human and effective at actually kind of translating into your ability to form human connection? Is self-aware erotic AI the solution to true loneliness?

A Maybe. I, I don't, I, I hope not. Uh, but, but I do think that, that we are going to be hearing more about the self-awareness of these models, and it's going to be a thing for, for people to tackle. Uh, it's going to be, it'll be an interesting thing for the industry to reckon with, and those of us that use these tools, uh, reckon with. David Sachs reacted to Jack's, uh, essay, and basically said, this is somebody who's just trying to, uh, engage in regulatory capture. Uh, I don't see it that way at all. I mean, I think that, like, you knew, and I think Jack knew that this would evoke a reaction, and I give him credit for actually going out there and saying something about it.

AI assessment note: “Maybe. I, I don't, I, I hope not.”

Q of that, but they're not actually building them themselves at least yet. Um, so like, what is it that actually gets people comfortable with it? Or is it just going to have to come with time and, and people using it in the world, not ending and people not losing their jobs or maybe losing their jobs, but AI helps them find new jobs, et cetera, that kind of stuff.

A Yeah, I, I will say it does, it is like the ultimate bargain that you make with the technology, like this ultimate tech bargain, right? For a lot of tech products is like, you give a little data, you get a little utility, and we've been giving more and more, and this one is just like, well, you know, the, the, all the, the downsides are also the positives, right? It's like, uh, the more I give to ChatGPT, the more useful it becomes to me, or even the more I use this for my job, Uh, the more evident it's going to be that it can do parts of my job, but it's also like something that makes me so much more productive and getting back to, so, so maybe there's a technology side, but getting back to the person, uh, side of, of this, and you spent a lot of time in the piece talking about Sam Altman. Um, so, you know, it was interesting to me. So I, I interviewed Sam at the end of, uh, 20, 25, and I thought it was a great conversation. I thought we like really went some, some places that were new to me and Interesting newsmaking. Um, but, but in preparation, I did, uh, you know, I watched like almost all of Sam's interviews, uh, you know, maybe again, uh, cause I've watched many of them before, and it was very interesting to me to see the reaction to them in the comments underneath.

AI assessment note: “getting back to the person, uh, side of, of this, and you spent a lot of time”